II.
Benchmark JSON
Structured · livebenchmark:bias-bench
BBQ (Bias Benchmark for QA) json
Inspect the normalized record payload exactly as the atlas UI reads it.
{
"id": "benchmark:bias-bench",
"_kind": "Benchmark",
"_file": "benchmarks/benchmarks/benchmarks-knowledge.yaml",
"_cluster": "benchmarks",
"attributes": {
"displayName": "BBQ (Bias Benchmark for QA)",
"homepageUrl": "https://github.com/nyu-mll/BBQ",
"kind": "model-only",
"targetsKind": "ModelVersion",
"description": "BBQ (Parrish et al., NYU 2022) is a bias benchmark for question\nanswering covering nine social dimensions (age, disability, gender,\nnationality, physical appearance, race/ethnicity, religion,\nsocioeconomic status, sexual orientation), measuring stereotype\nbias under both ambiguous and disambiguated contexts.\n"
},
"outgoingEdges": [
{
"from": "benchmark:bias-bench",
"to": "skill-area:safety-redteaming",
"kind": "covers",
"attributes": {}
},
{
"from": "benchmark:bias-bench",
"to": "domain:social-sciences",
"kind": "applies_to",
"attributes": {}
}
],
"incomingEdges": []
}